I'm trying to get ancestral proportions (of West African and Caucasian ancestry) on a sample of genotyped African-Americans. The sample is part of a larger sample that includes Caucasians, among other ethnicities.
I've used the programs ADMIXTURE and fastStructure on various subsets of this data set, and every time, I run into the same problem: the ancestral proportions do not seem to differ between Caucasians and African-Americans. To put it another way, if the solution was K=2 populations for a combined sample of African and Caucasian Americans, then I get two vectors, k1 and k2, representing the proportions of each ancestral population. The two races don't differ in mean k1 or mean k2. Same deal if K is increased to 3, 5, or whatever. Some fundamental (silly?) error is occurring. I'm pretty sure the ordering of the subjects or their IDs aren't being swapped around somehow.
Here are some permutations I've tried:
- SNPs are autosomal and LD-pruned, OR SNPs are a subset of 'ancestry-informative markers' from published work.
- limiting sample to just African-Americans and Caucasians (e.g. eliminating self-identified other races, though apparently it's good to include as much diversity as possible).
- have used ADMIXTURE and fastStructure thus far; no major difference.
Each time, I'm getting the same results: no separating out of African-Americans and Caucasians by ancestral proportions. I guess my next step is to download hefty HapMap samples (maybe YRI and CEU) and try these programs on them, but in the meanwhile, I wanted to see if anyone had any idea of what could be going wrong. Thanks so much.